WO2021088689A1 - Vehicle object detection - Google Patents

Vehicle object detection Download PDF

Info

Publication number
WO2021088689A1
WO2021088689A1 PCT/CN2020/124145 CN2020124145W WO2021088689A1 WO 2021088689 A1 WO2021088689 A1 WO 2021088689A1 CN 2020124145 W CN2020124145 W CN 2020124145W WO 2021088689 A1 WO2021088689 A1 WO 2021088689A1
Authority
WO
WIPO (PCT)
Prior art keywords
object detection
detection data
validated
video stream
vehicle
Prior art date
Application number
PCT/CN2020/124145
Other languages
English (en)
French (fr)
Inventor
Erik LINDBERG NILSSON
Jonathan JOHANSSON
Original Assignee
Ningbo Geely Automobile Research & Development Co., Ltd.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ningbo Geely Automobile Research & Development Co., Ltd. filed Critical Ningbo Geely Automobile Research & Development Co., Ltd.
Priority to EP20884471.2A priority Critical patent/EP4042324A4/de
Priority to CN202080075458.7A priority patent/CN114651285A/zh
Publication of WO2021088689A1 publication Critical patent/WO2021088689A1/en
Priority to US17/728,376 priority patent/US20220245951A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
    • G06F18/256—Fusion techniques of classification results, e.g. of results related to same input data of results relating to different input data, e.g. multimodal recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06V10/7747—Organisation of the process, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/809—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of classification results, e.g. where the classifiers operate on the same input data
    • G06V10/811—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of classification results, e.g. where the classifiers operate on the same input data the classifiers operating on different input data, e.g. multi-modal recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/76—Television signal recording
    • H04N5/765—Interface circuits between an apparatus for recording and another apparatus
    • H04N5/77—Interface circuits between an apparatus for recording and another apparatus between a recording apparatus and a television camera
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle
    • G06T2207/30261—Obstacle

Definitions

  • the present disclosure relates to the field of detection systems for vehicles. More particularly, it relates to a method, a computer program and a system for object detection training.
  • Object detections performed by vehicles are typically based on some kind of algorithm that has been tuned by means of real word data in order to give an as good detection performance as possible.
  • these algorithms involve manual analysation and labelling of the real word data, and tuning the algorithms may hence become cumbersome and high in costs.
  • American patent US 9158971 B2 describes a system and method for enabling generation of a specific object detector for a category of interest.
  • the method includes identifying seed objects in frames of a video sequence with a pre-trained generic detector for the category.
  • An appearance model is iteratively learned for each of the seed objects using other frames in which the seed object is identified.
  • An object of the present disclosure is to provide a method, a system and a computer program product where the previously mentioned problems are avoided or at least mitigated. This object is at least partly achieved by the features of the independent claims.
  • the physical product may comprise one or more parts, such as controlling circuitry in the form of one or more controllers, one or more processors, or the like.
  • a first aspect is a method for training an object detection system for a vehicle.
  • the object detection system comprises validated object detection data (VODD) of one or more objects.
  • the method comprises the steps of:
  • ⁇ identifying an object in the area by identifying one or more first object detection data of the object in the first video stream corresponding to validated object detection data
  • An advantage with the above aspect is that training of e.g. a detection algorithm of an object detection system for a vehicle becomes more reliable in terms of detecting objects since video streams from different vehicles featuring the same area are used for object detection training. Hence, if one vehicle has confirmed that it has detected e.g. a pedestrian, then the video streams of other vehicles covering the same area, but which may not be able to verify the object, may be used to update the object detection system in order to train the system to better recognize/detect objects.
  • object data which is gathered from the vehicles and which covers the same area and position in which an object has been validated i.e. if object data that is validated object detection data from one vehicle clearly able to identify the object at the same area and position exists, but which data from another vehicle does not unambiguously show the object, i.e. un-validated object detection data
  • object detection data that comprise un-validated object detection data i.e. it is not confirmed that the data actually portray the object
  • the system may be trained to recognize the object even if the image data is not complete, or the line of sight is obstructed.
  • the step of obtaining the first video stream comprising video images with object detection data of the area comprises receiving the first video stream from a first vehicle.
  • An advantage with the above embodiments is that live data for an area may be gathered easily.
  • the step of obtaining the second video stream comprising video images with object detection data of the area comprises receiving the second video stream from a second vehicle.
  • An advantage with the above embodiment is that data for an area may be gathered easily.
  • the object detection system may receive diverse data over the same area, which may be used for updating the system.
  • the diverse data provides better training and higher granularity compared to if data was only received from one vehicle.
  • the method may further comprise storing, at the second vehicle, the second video stream comprising video images with un-validated object detection data.
  • An advantage with the above embodiment is that data storage may be reduced since not all video streams have to be stored, but rather video streams that may be used as learning material (i.e. comprising video images with un-validated object detection data) may be stored locally and used at a later point in time.
  • the method may further comprise determining a time stamp of the first and second video streams and a vehicle location, wherein the time stamp indicating when the respective video stream was obtained, and wherein the vehicle location indicates a geographical position where the first and second video streams were obtained.
  • An advantage with the above embodiment is that only video streams that have been obtained/recorded at a valid point in time may be taken into consideration. I.e. vehicles that have video streams covering the area of the object, but which were recorded at a different point of time than a video stream having identified valid object detection data may in some embodiments not be taken into consideration.
  • Another advantage with the above embodiments is that the geographical location and/or orientation of the vehicle recording the video stream is determined and taken into account in order to easier determine the position of the object and in some embodiments to further determine which video streams should be taken into account for training the system.
  • determining whether the second object detection data is un-validated object detection data comprises correlating the second object detection data to validated object detection data and based on the correlation determining a confidence value of the second object detection data, wherein if the confidence value is determined to be below a confidence threshold, the second object detection data is determined to be un-validated object detection data.
  • An advantage with the above embodiments is that determining a correlation, e.g. a confidence value, between validated object detection data from the system or the first video stream and object detection data of the second video stream enables quick determination of whether the object detection data of the second video stream is valid or un-valid object detection data.
  • a correlation e.g. a confidence value
  • the validated object detection data that the second object detection data is correlated against is the validated object detection data of the first video stream.
  • An advantage of the above embodiment is that correlation is made between validated data covering the same area, and hence probably the same object is to be verified.
  • the update of the object detection system may thus be made based on the content of the first and the second video streams.
  • update the object detection system may mean update/train a detection algorithm of the object detection system so that it through self-learning can improve object detection.
  • the step of determining the position of the object comprises determining a distance and an angle to the object in relation to the first vehicle configured to obtain the first video stream comprising video images with object detection data of the area.
  • An advantage with the above embodiment is that a position of the detected object may be easily determined.
  • the step of identifying the object in the area further comprises identifying an object type of the object to be one or more of a person, a vehicle, fixed object, moving object or an animal.
  • An advantage with the above embodiments is that several different types of objects may be detected, such as persons being pedestrians or bicycle riders, Segway riders, kick bike riders, children, people in electric wheel chairs or vehicles such as trucks, other cars, trailers, motor bikes, and agricultural vehicles such as tractors and combines; fixed objects such as houses, rocks, trees, walls and signs; moving objects such as strollers, prams, wheel chairs, skate boards, shopping carts and lorries; or animals such as dogs, cats, horses, reindeers, wild hogs and rabbits.
  • objects such as persons being pedestrians or bicycle riders, Segway riders, kick bike riders, children, people in electric wheel chairs or vehicles such as trucks, other cars, trailers, motor bikes, and agricultural vehicles such as tractors and combines; fixed objects such as houses, rocks, trees, walls and signs; moving objects such as strollers, prams, wheel chairs, skate boards, shopping carts and lorries; or animals such as dogs, cats, horses, reindeers, wild hogs and rabbits.
  • these are just examples, other types
  • object type may relate to free space, such as background. I.e. when there is no object/object type to detect.
  • An advantage with detection free space is that the object detection system may train itself to determine when there actually is an object to detect, and when there is no object to detect.
  • the step of obtaining the second video stream comprising video images with object detection data of the area comprises identifying vehicles recording respective video streams comprising video images with object detection data of the area and requesting to receive the respective video streams.
  • An advantage of the above embodiment is that a larger amount of video data covering the area may be gathered. Hence, data covering different angles and distances of the same location may be used to train and update the object detection system leading to high granularity and a more reliable object detection.
  • the second video stream comprising video images with object detection data of the area comprises un-validated object detection data at the position of the object.
  • An advantage of the above embodiment is that video streams from other vehicles covering the desired location and area are preferably requested if they comprises detection data that is un-validated, i.e. it is not validated if the captured data actually comprise the validated object.
  • the un-validated object data may be correlated to the validated object data comprised in e.g. the first video stream, and it may hence be determined that the un-validated data is in fact validated data which may be used for training the system.
  • the correlation may e.g. be made through pattern recognition or by determining a confidence value denoting a match between the data or a probability that the data comprise the same object.
  • the control unit is configured to perform the steps of:
  • ⁇ identifying an object in the area by identifying one or more first object detection data of the object in the first video stream corresponding to validated object detection data
  • An advantage with the above aspect is that training of an object detection system for a vehicle becomes more reliable in terms of detecting objects since video streams from different vehicles featuring the same area are used for object detection. Hence, if one vehicle has confirmed that it has recorded e.g. a pedestrian, then the video streams of other vehicles recording the same area may be used to update the object detection system in order to provide further and different video images of the detected object.
  • Another advantage with the above embodiments is that by identifying and using object detection data that comprise un-validated object detection data (i.e. it is not confirmed that the data actually portray the object) , the system may be trained to recognize the object even if the video image data is not complete, or the line of sight is obstructed, since the system will know that it is confirmed that an object is at that location. Hence, a better training algorithm can be developed compared to if only confirmed images of the object were used for training.
  • the object detection system is comprised in a vehicle.
  • the object detection system is comprised in a remote server.
  • the object detection system comprises a system with several units.
  • the units may e.g. be comprised in vehicles and servers.
  • the system may be comprised only in vehicles.
  • control unit is configured to be connected to and receive video streams comprising video images with object detection data from at least a first and a second vehicle.
  • An advantage with the above embodiments is that data for an area may be gathered easily.
  • the object detection system may receive diverse data covering the same area, which may be used for updating the system.
  • the diverse data provides better training and higher granularity compared to if data was only received from one vehicle.
  • control unit is configured to store, at the second vehicle, the second video stream comprising video images with un-validated object detection data.
  • An advantage with the above embodiment is that data storage may be reduced since not all video streams have to be stored, but rather video streams that may be used as learning material (i.e. comprising un-validated object detection data) may be stored locally and used at a later point in time.
  • control unit is configured to determine a time stamp of the first and second video streams and a vehicle location, wherein the time stamp indicates when the respective video stream was obtained and wherein the vehicle location indicates a geographical location where the first and second video streams were obtained.
  • An advantage with the above embodiment is that only video streams that have been obtained/recorded at a valid point in time may be taken into consideration. I.e. vehicles that have video streams covering the area of the object, but which were recorded at a different point of time than a video stream having identified valid object detection data may not be taken into consideration.
  • Another advantage with the above embodiments is that the geographical location and/or orientation of the vehicle recording the video stream is determined and taken into account in order to easier determine the position of the object and in some embodiments to determine which video streams should be taken into account for training the system.
  • control unit is configured to determining whether the second object detection data is un-validated object detection data by correlating the second object detection data to validated object detection data and based on the correlation determining a confidence value of the second object detection data, wherein if the confidence value is determined to be below a confidence threshold, the second object detection data is determined to be un-validated object detection data.
  • An advantage with the above embodiments is that determining a correlation, e.g. a confidence value, between validated object detection data and object detection data of the video stream quick determination of whether the object detection data of the video stream is valid or un-valid object detection data may be done.
  • control unit is configured to identifying second object detection data comprising un-validated object detection data of the second video stream at the position of the object.
  • the step of identifying the object in the area further comprises identifying an object type of the object to be one or more of a person, a vehicle, fixed object, moving object or an animal.
  • An advantage with the above embodiments is that several different types of objects may be detected, such as persons being pedestrians or bicycle riders, Segway riders, kick bike riders, children, people in drive motors; or vehicles such as trucks, other cars, trailers, motor bikes, and agricultural vehicles such as tractors and combines; fixed objects such as houses, rocks, trees, walls and signs; moving objects such as strollers, prams, wheel chairs, skate boards, shopping carts and lorries; or animals such as dogs, cats, horses, reindeers, wild hogs and rabbits.
  • objects such as persons being pedestrians or bicycle riders, Segway riders, kick bike riders, children, people in drive motors; or vehicles such as trucks, other cars, trailers, motor bikes, and agricultural vehicles such as tractors and combines; fixed objects such as houses, rocks, trees, walls and signs; moving objects such as strollers, prams, wheel chairs, skate boards, shopping carts and lorries; or animals such as dogs, cats, horses, reindeers, wild hogs and rabbits.
  • these are just examples,
  • control unit is configured to identifying vehicles recording respective video streams comprising video images with object detection data of the area and request to receive the respective video streams.
  • An advantage with the above embodiments is that a larger amount of video data covering the area may be gathered. Hence, data covering different angles and distances of the same location may be used to train and update the object detection system leading to high granularity and a more reliable object detection.
  • the requested video streams comprises video images with object detection data of the area, the object detection data comprising un-validated object detection data associated with the position of the object.
  • An advantage with the above embodiment is that object data which is associated with the position of the object (i.e. the video stream has covered the position) , but which has not been validated to comprise validated object data (i.e. an object which should be detected such as a pedestrian) may be validated to comprise the object based on object data obtained from another video stream where the object has been validated.
  • object data which is associated with the position of the object (i.e. the video stream has covered the position) , but which has not been validated to comprise validated object data (i.e. an object which should be detected such as a pedestrian) may be validated to comprise the object based on object data obtained from another video stream where the object has been validated.
  • the training algorithm is given better granularity and more reliable object detection.
  • a third aspect is a computer program comprising instructions, which, when the program is executed by a computer, cause the computer to carry out the method according to the first aspect.
  • any of the above aspects may additionally have features identical with or corresponding to any of the various features as explained above for any of the other aspects.
  • Fig. 1 shows a flow chart illustrating method steps according to some embodiments
  • Fig. 2 shows schematically, an example detection scenario according to some embodiments
  • Fig. 3 shows block diagram of an example system according to some embodiments.
  • Fig. 4 shows a block diagram of an example computer program according to some embodiments.
  • Fig. 1 illustrates an example method 1 according to some embodiments.
  • the method 1 is for training an object detection system 10 for a vehicle 100, 31, 32, 33, 34.
  • the object detection system 10 comprises validated object detection data VODD of one or more objects, i.e. object data where it is confirmed that the object data portrays the one or more desired object to be detected.
  • the method 1 starts in step S1 with obtaining a first video stream VS1 comprising video images with object detection data ODD of an area 21.
  • Object detection data is made up of video images of the video stream VS1 capturing the immediate surroundings of a vehicle. Immediate surroundings may e.g. be a 5, 10, 50, 100, 400, 600 or more meter radius extending from the car.
  • step S2 the method continues with identifying S21 an object 22 in the area 21 in the immediate surroundings of the vehicle by identifying one or more first object detection data ODD1 of the object in the first video stream VS1 corresponding to validated object detection data VODD.
  • Validated object detection data may e.g. be stored in the object detection system and comprise a database of validated objects.
  • Validated objection data may e.g. correspond to a multitude of images portraying varying objects in varying settings, which helps the object detection system to learn and recognize (i.e. be trained for detection) and hence detect objects captured in the video stream.
  • validated object detection data is a detection algorithm analysing (e.g. by applying pattern recognition) the content of the object detection data of the video stream in order to determine whether the content comprise e.g. pixels forming an object that should be detected.
  • the obtained first object detection data may be correlated against the validated object detection data of the object detection system.
  • the correlation may e.g. comprise comparing the video image content of the first video stream to or with the valid object detection data and determine a correlation result or confidence value for indicating a probability that the first object detection data is valid object detection data. If the correlation value or confidence result indicates that there is e.g. a higher than 60%probability that the first object detection data comprise an video image of e.g. a person, the first object detection data may be labeled, or determined as valid object detection data. It should be noted that 60%is just an example, and values ranging both higher and lower are contemplated.
  • determining a position 23 of the object 22 may also comprise tagging the position 23 with a time stamp, i.e. determining a time stamp of the video stream.
  • the time stamp may enable the object detection system 10 to collect data collected within a predetermined time range. For example, for moving objects such as pedestrians, cyclers, vehicles etc. there may be little point in collecting video streams over the same area that are captured hours or days later. However, in some scenarios where objects have been detected to be at a validated location for longer periods of time (e.g. if the object is fixed, or if a pattern of movement has been detected such as a commuter being detected at the same area approximately the same time every day) it may be beneficial to collect video streams covering the area for longer periods of time as well.
  • determining the position 23 of the object 22 may alternatively or additionally comprise determining a vehicle location and/or vehicle orientation of the vehicle recording the respective video stream.
  • determining the position 23 of the object 22 may alternatively or additionally comprise determining a vehicle location and/or vehicle orientation of the vehicle recording the respective video stream.
  • step S4 the method 1 comprises obtaining a second video stream VS2 comprising video images with object detection data ODD of the area 21.
  • object detection data is made up of video images capturing the immediate surroundings of the vehicle.
  • a time stamp is determined for both the first and the second video stream.
  • step S5 the method comprises identifying second object detection data ODD2 of the second video stream VS2 at the position 23 of the object 22 and determining whether the second object detection data ODD2 is un-validated object detection data UODD.
  • the method 1 may comprise that the second video stream is analysed such that video images comprising the position of the object is taken into account.
  • the step S51 of determining whether the second object detection data ODD2 is un-validated object detection data UODD may comprise correlating the second object detection data ODD2 to validated object detection data VODD and based on the correlation determining a confidence value of the second object detection data ODD2. If the confidence value is determined to be below a confidence threshold, the second object detection data ODD2 is determined to be un-validated object detection data UODD.
  • the method 1 may comprise identifying or determining that at least one vehicle 100, 31, 32, 33, 34 (e.g. the second vehicle) is not being able to verify that what is seen in the object detection data ODD of the second video stream VS2, at the position 23 of the object 22, is actually the object 22.
  • the second video stream VS2 comprise un-validated object detection data UODD associated with the position 23 of the object 22.
  • the second object detection data ODD2 may be correlated against the validated object detection data VODD of the first video stream VS1.
  • the validated object detection data VODD of the first stream VS1 may be used in order to determine whether the second object detection data ODD2 is valid or un-valid object detection data. If the second object detection data ODD2 is determined to be un-valid object detection data UODD based on the correlation with the first video stream VS1, the object detection system 10 may use the un-valid object detection data UODD of the second stream VS2 in order to train itself to find a pattern and better recognize objects. This may be enabled since the first video stream VS1 comprise validated object detection data VODD over an object 22 at a determined position. The second video stream VS2 covering the same area 21 and position 23 should hence also see and be able to verify the object 22.
  • the second video stream VS2 may for some reason comprise inferior quality or be partially obstructed and only parts of the object are discernible, but not enough to perform a correlation or pattern detection that results in valid object detection. In such case, the second video stream VS2 still possibly show the object 22, but the algorithm of the object detection system 10 is not able to verify it.
  • the object detection system 10 may use the un-validated object detection data UODD for training itself to recognize objects based on the fact that the first video stream VS1 comprises validated object detection data VODD of the object 22, and the second video stream VS2 should possibly hence as well.
  • un-validated data may be determined as un-validated data simply because the object may have moved rapidly out of the way, or walked behind a tree or is obscured by a passing vehicle, etc. and a second vehicle is simply not seeing it.
  • the un-validated data may still be used for training the system. It may e.g. be beneficial for the algorithm to learn how do recognize background data (free space) , i.e. scenes where there is no object to detect.
  • the object detection system may train itself to recognize both objects and non-objects based on un-validated object detection data (which data may comprise both un-validated objects and non-existing objects.
  • the method continues in step S6 with updating the validated object detection data VODD of the object detection system 10 if it is determined that the second object detection data ODD2 of the object 22 is un-validated object detection data UODD.
  • the method may also comprise updating the validated object detection data VODD with the un-validated object detection data UODD based on the validated object detection data VODD of the first video stream VS1.
  • the step S1 of the method 1 comprising obtaining the first video stream VS1 comprising video images with object detection data ODD of the area may optionally further comprise receiving in step S11 the first video stream VS1 from a first vehicle.
  • the object detection system 10 may be located in one or more vehicles 100, 31, 32, 33, 34 but may also communicate wirelessly with and/or comprise a server 200 in e.g. a network cloud.
  • the server 200 may e.g. collect video streams from the first vehicle 100, 31 as well as from other vehicles 100, 31, 32, 33, 34 and perform the training of the object detection system 10 based on the received video streams and the object detection data ODD comprised therein.
  • the server 200 may then update the validated object detection data VODD (e.g. an object detection algorithm) and push this update through the network cloud to the object detection system 10 of each respective vehicle 100, 31, 32, 33, 34.
  • VODD e.g. an object detection algorithm
  • step S4 of the method 1 comprising obtaining the second video stream VS2 comprising video images with object detection data ODD of the area may optionally further comprise receiving in step S41 the second video stream VS1 from a second vehicle 100, 32, 33, 34.
  • the server as described above may obtain video streams from a second vehicle.
  • the first vehicle 100, 31 may be configured to obtain the second video VS2 stream from the second vehicle 100, 32, 33, 34.
  • the first vehicle may transfer the first; the first and the second; or only the second video stream to the server for update of the object detection system.
  • the second vehicle may obtain the first video stream VS1 from the first vehicle and may transfer the first video stream; the second video stream; or both the first and second video stream to the server.
  • the object detection system of the respective first and second vehicle may perform the object detection data update locally in each vehicle without involving an external server 200.
  • the second vehicle may further store the second video stream VS2 comprising un-validated object detection data UODD.
  • the video stream comprising video images with un-validated object detection data UODD locally at the second vehicle (or at the vehicle that captured the video stream comprising video images with un-validated object detection data) data storage and storage space may be improved.
  • the vehicles and/or server may focus on storing video streams that comprise un-validated object detection data in favour of storing validated object detection data. The stored streams may then be used for training the algorithm/object detection system at a later stage.
  • the step of determining S3 the position of the object in the method 1 may optionally comprise the step S31 comprising determining a distance D1, D2, D3 and an angle ⁇ 1, ⁇ 2, ⁇ 3 to the object in relation to the first vehicle configured to obtain the first video stream VS1 comprising video images with object detection data ODD of the area.
  • the method 1 may enable a vehicle to determine the location 23 of a detected object 22, which location 23 may then be used when gathering and analysing other video streams of the area 21 in order to detect or not detect the object 22 in other video streams and thus train the object detection system to recognize un-validated object detection data UODD as validated object detection data VODD.
  • the step S2 in method 1 of identifying (step S21) the object in the area may optionally further comprises the step S22 of identifying an object type of the object 22 to be one or more of a person, a vehicle, fixed object, moving object or an animal.
  • the object type By identifying the object type, restrictions can be made in how much object detection data should be gathered and for how long. If the object type is determined to be a person, a vehicle, moving object or an animal then it may be of greater interest to collect video streams in a shorter period time than if the type of object is determined to be a fixed object. Furthermore, for fixed objects it may be of more interest to focus on gathering un-validated object detection data in order to train the object detection system to better recognize the object. For example, if the first vehicle determines that it has detected the fixed object, then its video stream comprising video images with the first object detection data may not have much value for training the system.
  • the second vehicle comes to the area where there is supposed to be a fixed object that should be detected, but fails to detect it in the second video stream even though the video stream is covering the area and the supposed position of the object, then that second video stream comprising video images with the un-validated object detection data may be of greater interest when training the system.
  • the step S4 of the method 1 comprising obtaining the second video stream VS2 comprising video images with object detection data ODD of the area may optionally comprise the step S42 of identifying vehicles 100, 31, 32, 33, 34 recording respective video streams VS comprising video images with object detection data ODD of the area and requesting to receive the respective video streams VS.
  • Recording, and in some embodiments requesting may further be comprised in optional step S7 of the method 1.
  • the recording and requesting may comprise two different method steps, e.g. step S7 of recording and step S8 (not shown in Fig. 1) of requesting.
  • the second video stream VS2 comprising video images with object detection data ODD of the area 21 comprises un-validated object detection data UODD at the position 23 of the object 22.
  • the object detection data of the second video stream may e.g. comprise blurry video images, or partial video images which leads to that it cannot be confirmed whether the object detection data associated with the position of the object actually shows the object, and the object detection data hence comprise un-validated object detection data at the position of the object.
  • the method 1 has been described as being performed in a series of method steps in a validated order. It should be noted that the order of the steps may in some embodiments be another than that described above. For example, in some embodiments, the steps S4 and S5 may switch place with the steps of S1 and S2.
  • the method 1 as described above defines a scenario where a first vehicle has detected an object and the object detection system of the first vehicle has validated it as validated object detection data. I.e. the object detection system of the first vehicle has validated that is has detected e.g. a person, vehicle, fixed object, animal etc. and may ask other vehicles in the area whether they see the same. If the other vehicles in the area determines that they do not, i.e. their object detection data associated with the position of the object and comprised in their respective videos stream is un-validated object detection data, then their video streams comprising video images with the un-validated object data may be used to update validated object detection data and hence train the system.
  • a vehicle may obtain a video stream of an area, the video stream comprising video images with object detection data.
  • the object detection system of the vehicle may react to that something is present in the object detection data, but it cannot be verified what it is.
  • the video stream may e.g. be of inferior quality because of weather conditions (rain may e.g. result in blurred or inferior video images that are hard to interpret) , or the video images are partly obstructed, or are blurry or for any other reasons do not provide object detection data that can be matched to validated object detection data.
  • the system may determine that a confidence value of the object detection data indicates a 42%probability that the data shows a person.
  • the vehicle (or the object detection system comprised in the vehicle) may then inquire with other vehicles in the area if their video streams have captured validated object detection data associated with the location of the un-validated object detection data.
  • the vehicle may receive the video streams from the other vehicles that comprise validated detection data, and may then update the object detection system based on the obtained/received un-validated and validated object detection data.
  • the vehicle (s) may transmit their respective video streams to an external server in order to update the object detection system.
  • Fig. 2 illustrates an example scenario where the method and embodiments described above may be applicable.
  • a first vehicle 31, a second vehicle 32, a third vehicle 33 and a fourth vehicle 34 are present on a road.
  • the first vehicle 31 may e.g. be the first vehicle as described in conjunction with Fig. 1.
  • the second 32, third 33 and fourth 34 vehicle may be the second vehicle as described in conjunction with Fig. 1.
  • the vehicles 31, 32, 33 and 34 are illustrated as cars, this should be seen merely as an example as other types of vehicles are also possible, such as trucks, motor bikes, recreational vehicles, busses, etc.
  • the first 31, second 32, third 33 and 34 fourth vehicle are all equipped with a respective object detection system 10.
  • the object detection system 10 of each respective vehicle provides obtaining a respective video stream VS1, VS2, VS3 and VS4 (VS4 is not shown in Fig. 2 for reasons that will be explained below) , of an area 21.
  • an object 22 is present at a position 23.
  • the object 22 is illustrated as a bicyclist in Fig. 2, this is however just an example (for simplicity, all though a bicycle also is considered as a vehicle, the cyclist of Fig. 2 is not denoted as vehicle but as the object in this disclosure) .
  • the object 22 could also be any other type of object such as a pedestrian/person, vehicle, fixed object or animal, as described in conjunction with Fig. 1.
  • the object 22 could furthermore be situated at another location than in the middle of road, such as on the pavement or similar.
  • the object detection system 10 of the vehicles may be configured to transmit the recorded video streams VS1-VS4 to an external server 200 illustrated as a cloud in Fig. 2.
  • the video streams illustrated in Fig. 2 are exemplary, and may have other ranges.
  • the video streams could cover a semicircle spanning 180 degrees from the object detection systems.
  • the range and coverage of the video streams may be dictated by the type of unit that is recording the streams. Some cameras may e.g. record a full circle of 360, degrees, other may record a part of a circle of e.g. 270, 180, 90, 60 etc. degrees. It should also be noted that circular ranges are an example and other shape of ranges are contemplated.
  • the length of the ranges may also vary according to the limitations of the recording camera/unit.
  • the object detection system 10 of the first vehicle 31 may thus obtain the first video stream VS1 comprising video images with object detection data of the area 21.
  • the object detection system 10 of the first vehicle 31 may then identify the object 22 in the area 21 by identifying one or more first object detection data ODD1 in the first video stream VS1 corresponding to validated object detection data VODD.
  • the one or more first object detection data ODD1 may e.g. clearly show the person on the bicycle (i.e. the cyclist) in the road.
  • the object detection system 10 of the first vehicle 31 may comprise a database of validated object detection data, and when comparing the one or more first object detection data ODD1 to the validated object detection data there is a clear match and the object detection system of the first vehicle may then determine that it sees/has detected an object 22.
  • the validated object detection data of the object detection system may alternatively or additionally be an algorithm instructing the object detection system what to look for in object detection data in order to determine whether the object detection data is valid or not.
  • the algorithm may e.g. comprise a series of patterns that should be fulfilled when analysing the pixels of the video streams in order to determine valid or un-valid object detection data.
  • the object detection system 10 of the first vehicle may determine the position 23 of the object 22 in the area 21.
  • the first detection system may e.g. determine a distance D1 and an angle ⁇ 1 to the object 22 in relation to the first vehicle 31.
  • the object detection system 10 may further tag the first video stream VS1 of the area 21 and the object 22 with a time stamp.
  • the object detection system 10 may be configured to determining a time stamp of the first and second video streams, the time stamp indicating when the respective video stream was obtained.
  • the object detection system 10 may be configured to determining a vehicle location and/or orientation of the vehicle recording the respective video stream.
  • the video stream may be tagged (in addition or alternatively to the time stamp) with the vehicle location.
  • the vehicle location may be a geographical location denoting the physical position of the vehicle and may be determined by means of e.g. GPS.
  • the object detection system 10 of the first vehicle may inquire other vehicles in the area, e.g. the second 32, third 33 and fourth 34 vehicles if they have a video stream over the area, and if they see the object 22.
  • the object detection system of the first vehicle 31 may e.g. identify other vehicles recording video streams comprising video images with object detection data of the area.
  • the second 32 and third 33 vehicle may respond whereas the fourth vehicle 34 may not since it does not capture the area 21 (and its video stream VS4 is hence not shown in Fig. 2) .
  • Whether or not the fourth vehicle should respond may be dictated by the time stamp. If e.g. only real time video streams are of interest, the video stream of the fourth vehicle is not of interest since it does not capture the area 21 at the required time.
  • the second 32 and third 33 vehicles may respond only if they determine the object detection data of each respective video stream VS2 and VS3 to comprise un-validated object detection data UODD associated with the position 23 of the alleged object 22 in the area 21.
  • the second and third vehicles may try to identify the object 22 by e.g. also determine a distance D2, D3 and angle ⁇ 2, ⁇ 3 to the alleged object 22 in relation to the second and third vehicles respectively, and determine if the object detection data associated with the position is un-validated UODD or validated object detection data VODD.
  • the second and/or third object detection data ODD2, ODD3 of the second and/or third video streams VS2, VS3 is determined to be un-validated object detection data UODD
  • the second and/or third video streams comprising video images with un-validated object detection data UODD may be used for updating and thereby training the object detection system.
  • the object detection system 10 of the first vehicle may obtain all video streams of the area 21 (in Fig. 2, the VS2 and VS3) and identify whether the comprised second and third object detection data ODD2, ODD3 of VS2 and VS3 comprise the object 22 by trying to identify the object 22 at the position 23 in the area 21 in the video streams VS2 and VS3.
  • the object detection system 10 of the first vehicle may further update validated object detection data with the object detection data obtained from the second and/or third video stream.
  • the video streams VS1, VS2 and VS3 may be obtained by the external server 200 (e.g. a server in the cloud, as shown in Fig. 2) over a network connection such as the internet.
  • the external server may perform the update of the validated object detection data if it is determined that obtained un-validated object detection data UODD possibly is correlated with validated object detection data VODD of the area, e.g. from the first video stream, (and thus train e.g. the algorithm of the object detection system that performs the detection) and then push the update to the object detection systems 10 of the first, second, third and fourth vehicles so that the respective object detection system 10 may be trained to recognize objects in various settings.
  • the object detection system is updated based on a first vehicle detecting validated object detection data (i.e. the first vehicle knows what object it is seeing) and other un-validated object detection data is then gathered from vehicles that cannot verify that they see the object in the same position.
  • the gathered data is then used to update/train the system.
  • a blurry video image, video image with inferior resolution or partial video image of the object i.e. un-validated object detection data
  • the first vehicle may detect a pedestrian at a distance of 20 m. Detection is certain and is based on a video image of the pedestrian having a size of e.g. 200*50 pixels.
  • the second vehicle may be further away and sees the same area but at a 250 m distance.
  • the pedestrian may in such case be captured by 10*3 pixels which gives less good resolution than what the first vehicle could obtain and it may hence be more difficult for the second vehicle to know what it’s seeing and the video stream of the second vehicle is hence valuable for training the system.
  • the video images may e.g. compared to each other and details matched to confirm that the un-validated object detection data from the second vehicle is in fact validated object detection data and determine e.g. a pattern in the un-validated object detection data that may be used in the future to determine whether object detection data is validated or un-validated.
  • the updated/trained object detection system may determine that the video image is validated object detection data based on the updated training algorithm of the object detection system.
  • the first vehicle may e.g. be the second, third or fourth vehicle and vice versa.
  • the method may start with a vehicle determining that it cannot validate that what it is actually seeing in its video stream of an area is a validated object.
  • the third vehicle 33 may detect un-validated object detection data UODD in its video stream VS3 at the position 23 of the area 21.
  • the video stream VS3 may e.g. comprise a partially obstructed video image of the object 22.
  • the line of sight to the object 22 from the third vehicle is e.g. partially obstructed by the second vehicle 32.
  • a confidence value of the object detection data of the video stream may e.g. be 40%which may not be high enough to pass a threshold for validated object detection data, but still be high enough such that the object detection system determines that there might be an object in the video stream that should be detected.
  • the third vehicle 33 may then after having determined that the third video stream VS3 comprise un-validated object detection data UODD associated with/at the position 23 of the area 21, identify other vehicles that have recorded a respective video stream covering the area 21.
  • the third vehicle 33 may e.g. send out an enquiry to other present vehicles if they have detected validated object detection data associated with the position.
  • the second 32 and first 31 vehicle may respond by transmitting their respective video stream comprising video images with validated object detection data of the object 22 to the third vehicle 33.
  • the third vehicle may then update the object detection system based on the determined un-validated object detection data and possibly the validated object detection data.
  • the third vehicle 33 may locally update its object detection system 10, and possibly transmit the update to the other vehicles such that their respective object detection systems 10 are updated as well.
  • the third vehicle may transfer the obtained un-validated and possibly the validated object detection data to an external server comprising an object detection system and a database of validated object detection data and/or algorithms for recognizing validated object detection data.
  • the external server may then use the obtained data to train/update the object detection system and possibly push the update to all object detection systems connected to the server and associated with a vehicle (e.g. the respective object detection system 10 associated with vehicles 31, 32, 33 and 34) .
  • Figure 3 illustrates in a block diagram an object detection system 10 for a vehicle 100 according to some embodiments.
  • the object detection system 10 may e.g. be the object detection system as described in conjunction with any of the previous Figs. 1-2.
  • the vehicle 100 may e.g. be any of the vehicles as described in conjunction with Figs. 1-2.
  • the object detection system 10 comprises a control unit 11 (CNTR) and an object detection data module 112 (ODD) comprising validated object detection data 113 (VODD) and un-validated object detection data 114 (UODD) of one or more objects.
  • CNTR control unit 11
  • ODD object detection data module 112
  • VODD validated object detection data 113
  • UODD un-validated object detection data 114
  • control unit 11 may comprise controlling circuitry.
  • the control unit/controlling circuitry may comprise the object detection data module 112 for storing object detection data/algorithms of validated objection data 113 and un-validated object detection data 114.
  • control unit may further comprise a video unit 111 (VID) , and a determining unit 115 (DET) .
  • the object detection system 10 may further comprise an antenna circuit 12 (RX/TX) .
  • the control unit 11 is configured to perform obtaining of a first video stream (e.g. the VS1 of Fig. 2) comprising video images with object detection data of an area (compare with step S1 of the method 1) .
  • the control unit 11 may e.g. be configured to cause the video unit 111 to record and relay a first video stream VS1 and cause the object detection data module 112 to store the object detection data ODD of the first video stream VS1.
  • the control unit may further be configured to cause identifying of an object 22 in the area 21 by identifying one or more first object detection data ODD1 of the object 22 in the first video stream VS1 corresponding to validated object detection data VODD (compare with step S2 of the method 1) .
  • the control unit 11 may e.g. be configured to cause the ODD module 112 to determine whether the obtained object detection data corresponds to validated objection data or un-validated object detection data e.g. by using an algorithm for object detection and storing the object detection data as either validated object detection data 113 or un-validated object detection data 114.
  • the control unit 11 may be configured to cause determining of a position 23 in the area 21 of the object 22 (compare with step S3 in the method 1) .
  • the control unit 11 may e.g. be configured to cause the determining module 115 to determine the position 23.
  • the control unit 11 may be configured to cause obtaining of a second video stream VS2 comprising video images with object detection data of the area (compare to step S4 in method 1) .
  • the control unit 11 may e.g. be configured to cause the antenna circuit 12 to receive the second video stream VS2.
  • the control unit 11 may be configured to cause identifying of second object detection data ODD2 of the second video stream VS2 at the position 23 of the object 22, (compare with step S5 of method 1) and determine whether the second object detection data ODD2 is un-validated object detection data UODD.
  • the control unit 11 may e.g. cause the object data detection module 112 to analyse the second object detection data ODD2 by means of stored algorithms for validated object detection data VODD, and/or match second object detection data ODD2 to stored validated object detection data VODD or to the validated object detection data VODD of the first video stream VS1.
  • the control unit 11 may e.g.
  • the object detection module 112 may be configured to cause the object detection module 112 to determine whether the second object detection data ODD2 matches to validated detection data VODD stored in the module 112, 113. When no match is determined, the second object detection data ODD2 may be seen and identified as un-validated detection data UODD at the position 23 of the object 22.
  • control unit 11 may be configured to cause determining whether the second object detection data ODD2 is un-validated object detection data VODD by correlating the second object detection data ODD2 to validated object detection data VODD and based on the correlation determining a confidence value of the second object detection data, wherein if the confidence value is determined to be below a confidence threshold, the second object detection data is determined to be un-validated object detection data (compare with step S51 of method 1) .
  • the control unit 11 may be configured to cause updating of the validated object detection data 113 of the object detection system 10 if it is determined that the second object detection data of the object 22 is un-validated object detection data UODD.
  • the control unit 11 may e.g. cause the object detection module 112 to store the second objection detection data in the validated object detection database 113 as validated objection detection data and/or update a stored detection algorithm.
  • control unit 11 is configured to be connected to and receive video streams comprising video images with object detection data from at least a first and a second vehicle 100, 31, 32, 33, 34 (compare with method 1 and Fig. 2) .
  • the control unit 11 may e.g. be configured to cause the video module 111 to record a video stream, and/or cause the antenna circuit 12 to receive one or more video streams from at least a first and a second vehicle.
  • control unit 11 is further configured to store at the second vehicle 100, 32, 33, 34 the second video stream VS2 comprising un-validated object detection data UODD.
  • control unit 11 is further configured to determine a time stamp of the first and second video streams VS1, VS2, the time stamp indicating when the respective video stream was obtained.
  • control unit 11 is configured to identifying an object type of the object to be one or more of a person, a vehicle, fixed object or an animal (compare with step S22 of method 1) .
  • the control unit 11 may e.g. cause the ODD module 112, possibly in cooperation with the determining module 115 to determine and thereby identify (based on the object detection data) an object type of the object.
  • control unit 11 is configured to identifying vehicles recording video streams comprising video images with object detection data of the area (compare with steps S42 and S7 in method 1) .
  • the control unit 11 may for example be configured to cause antenna circuit 12 to search for and identify other vehicles in the area.
  • control unit 11 is configured to identifying vehicles recording video streams comprising video images with object detection data ODD of the area 21, the object detection data ODD comprising un-validated object detection data UODD at the position 23 of the object 22 (compare with steps S42 and S7 of method 1) .
  • the control unit 11 may e.g. be configured to cause the object detection module 112 to determine that the object detection data ODD is un-validated object detection data UODD.
  • the object detection system 10 as described in Fig. 3 may be comprised in an external server 200.
  • the object detection system 10 may be configured to communicate with other object detection systems 10 comprised in vehicles 100, 31, 32, 33, 34 and (wirelessly) connected to the external server 200.
  • Fig. 4 illustrates a computer program comprising instructions, which, when the program is executed by a computer, cause the computer to carry out the methods as described in conjunction with any of the previous figs. 1-3.
  • FIG. 4 illustrates in some embodiments a computer program product on a non-transitory computer readable medium 400.
  • Figure 4 illustrates an example non-transitory computer readable medium 400 in the form of a compact disc (CD) ROM 400.
  • the non-transitory computer readable medium has stored thereon a computer program comprising program instructions.
  • the computer program is loadable into a data processor (PROC; e.g., data processing circuitry or a data processing unit) 420, which may, for example, be comprised in a control unit 410.
  • PROC data processing circuitry or a data processing unit
  • the computer program may be stored in a memory (MEM) 430 associated with or comprised in the data processor.
  • the computer program may, when loaded into and run by the data processor, cause execution of method steps according to, for example, any of the methods illustrated in Figures 1-3, or otherwise described herein.
  • a non-transitory computer-readable storage medium storing one or more programs configured to be executed by one or more processors of a system for object detection, the one or more programs comprising instructions for performing the method according to any one of the above-discussed embodiments.
  • a cloud computing system can be configured to perform any of the method aspects presented herein.
  • the cloud computing system may comprise distributed cloud computing resources that jointly perform the method aspects presented herein under control of one or more computer program products.
  • the processor may be connected to one or more communication interfaces and/or sensor interfaces for receiving and/transmitting data with external entities such as e.g. sensors arranged on the vehicle surface, an off-site server, or a cloud-based server.
  • the processor (s) may be or include any number of hardware components for conducting data or signal processing or for executing computer code stored in memory.
  • the system may have an associated memory, and the memory may be one or more devices for storing data and/or computer code for completing or facilitating the various methods described in the present description.
  • the memory may include volatile memory or non-volatile memory.
  • the memory may include database components, object code components, script components, or any other type of information structure for supporting the various activities of the present description. According to an exemplary embodiment, any distributed or local memory device may be utilized with the systems and methods of this description.
  • the memory is communicably connected to the processor (e.g., via a circuit or any other wired, wireless, or network connection) and includes computer code for executing one or more processes described herein.
  • ODD1 First object detection data
  • ODD2 Second object detection data
  • ODD3 Third object detection data

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Signal Processing (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)
PCT/CN2020/124145 2019-11-06 2020-10-27 Vehicle object detection WO2021088689A1 (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP20884471.2A EP4042324A4 (de) 2019-11-06 2020-10-27 Fahrzeugobjekterkennung
CN202080075458.7A CN114651285A (zh) 2019-11-06 2020-10-27 车辆对象探测
US17/728,376 US20220245951A1 (en) 2019-11-06 2022-04-25 Vehicle object detection

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
EP19207494.6 2019-11-06
EP19207494.6A EP3819811A1 (de) 2019-11-06 2019-11-06 Fahrzeugobjekterkennung

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/728,376 Continuation US20220245951A1 (en) 2019-11-06 2022-04-25 Vehicle object detection

Publications (1)

Publication Number Publication Date
WO2021088689A1 true WO2021088689A1 (en) 2021-05-14

Family

ID=68470414

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/124145 WO2021088689A1 (en) 2019-11-06 2020-10-27 Vehicle object detection

Country Status (4)

Country Link
US (1) US20220245951A1 (de)
EP (2) EP3819811A1 (de)
CN (1) CN114651285A (de)
WO (1) WO2021088689A1 (de)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115139999B (zh) * 2022-06-28 2023-09-15 重庆长安汽车股份有限公司 车辆与行人防撞控制方法及装置、电子设备、存储介质
EP4358038A1 (de) * 2022-10-18 2024-04-24 Volvo Car Corporation Verfahren zum erkennen eines objektes im umfeld eines fahrzeugs, verfahren zum steuern eines fahrzeugs, objekterkennungssystem, datenverarbeitungsvorrichtungen, computerprogramme und computerlesbare speichermedien

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140072173A1 (en) * 2012-09-12 2014-03-13 International Business Machines Corporation Location determination for an object using visual data
US20150286865A1 (en) * 2014-04-08 2015-10-08 Sony Corporation Coordination of object location data with video data
US20180373943A1 (en) * 2017-06-23 2018-12-27 Panasonic Intellectual Property Corporation Of America Computer implemented detecting method, computer implemented learning method, detecting apparatus, learning apparatus, detecting system, and recording medium
CN109479118A (zh) * 2016-09-30 2019-03-15 富士通株式会社 对象检测方法、对象检测装置以及电子设备
EP3466762A1 (de) * 2017-10-05 2019-04-10 Ningbo Geely Automobile Research & Development Co. Ltd. Überwachungssystem und verfahren zur anzeige eines oder mehrerer teilbilder für einen benutzer eines fahrzeugs

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9158971B2 (en) 2014-03-03 2015-10-13 Xerox Corporation Self-learning object detectors for unlabeled videos using multi-task learning
DE102017215552A1 (de) * 2017-09-05 2019-03-07 Robert Bosch Gmbh Plausibilisierung der Objekterkennung für Fahrassistenzsysteme
US11232350B2 (en) * 2017-11-29 2022-01-25 Honda Motor Co., Ltd. System and method for providing road user classification training using a vehicle communications network
EP3547215A1 (de) * 2018-03-26 2019-10-02 Cohda Wireless Pty Ltd. Systeme und verfahren zum automatischen trainieren neuronaler netzwerke

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140072173A1 (en) * 2012-09-12 2014-03-13 International Business Machines Corporation Location determination for an object using visual data
US20150286865A1 (en) * 2014-04-08 2015-10-08 Sony Corporation Coordination of object location data with video data
CN109479118A (zh) * 2016-09-30 2019-03-15 富士通株式会社 对象检测方法、对象检测装置以及电子设备
US20180373943A1 (en) * 2017-06-23 2018-12-27 Panasonic Intellectual Property Corporation Of America Computer implemented detecting method, computer implemented learning method, detecting apparatus, learning apparatus, detecting system, and recording medium
EP3466762A1 (de) * 2017-10-05 2019-04-10 Ningbo Geely Automobile Research & Development Co. Ltd. Überwachungssystem und verfahren zur anzeige eines oder mehrerer teilbilder für einen benutzer eines fahrzeugs

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4042324A4 *

Also Published As

Publication number Publication date
EP3819811A1 (de) 2021-05-12
EP4042324A4 (de) 2024-01-03
CN114651285A (zh) 2022-06-21
EP4042324A1 (de) 2022-08-17
US20220245951A1 (en) 2022-08-04

Similar Documents

Publication Publication Date Title
US11093801B2 (en) Object detection device and object detection method
CN109902552B (zh) 用于车道特征分类和定位的人工神经网络
US20220245951A1 (en) Vehicle object detection
US10657391B2 (en) Systems and methods for image-based free space detection
US20200019165A1 (en) System and method for determining a vehicles autonomous driving mode from a plurality of autonomous modes
US9542609B2 (en) Automatic training of a parked vehicle detector for large deployment
US20190092318A1 (en) Systems and methods for rear signal identification using machine learning
US20200250440A1 (en) System and Method of Determining a Curve
US20200097739A1 (en) Object detection device and object detection method
EP3007099A1 (de) Bilderkennungssystem für ein Fahrzeug und entsprechendes Verfahren
US11170272B2 (en) Object detection device, object detection method, and computer program for object detection
US20080166018A1 (en) Method and apparatus for performing object recognition on a target detected using motion information
CN111508260A (zh) 车辆停车位检测方法、装置和系统
WO2015189847A1 (en) Top-down refinement in lane marking navigation
KR101678004B1 (ko) 노드-링크 기반 카메라 네트워크 통합 감시 시스템 및 감시 방법
WO2015177864A1 (ja) 信号機認識装置及び信号機認識方法
US10929715B2 (en) Semantic segmentation using driver attention information
WO2021062596A1 (en) Systems and methods for predicting a vehicle trajectory
CN111027381A (zh) 利用单目相机识别障碍物的方法、装置、设备及存储介质
US11756303B2 (en) Training of an object recognition neural network
CN110658539A (zh) 车辆定位方法、装置、车辆和计算机可读存储介质
CN114663852A (zh) 车道线图的构建方法、装置、电子设备及可读存储介质
CN112883058A (zh) 用于车辆定位的标定方法、装置、设备、车辆和介质
Nambi et al. ALT: towards automating driver license testing using smartphones
JP2021128705A (ja) 物体状態識別装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20884471

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020884471

Country of ref document: EP

Effective date: 20220510

NENP Non-entry into the national phase

Ref country code: DE